Project

General

Profile

Actions

Bug #726

open

Incorrect management of internal dust extinction in disks?

Added by Emmanuel Bertin about 5 years ago. Updated about 5 years ago.

Status:
In Progress
Priority:
High
Start date:
10/06/2016
Due date:
10/10/2016 (over 5 years late)
% Done:

70%

Estimated time:
1.00 h
Spent time:

Description

(!) This is an issue previously reported by Damien Leborgne, which I erased by mistake...

In stuff version 1.26
---------------------

Unwanted or strange behaviour:
if a population is defined with B/T (BULGE_FRACTION) < 1 and DISK_EXTINCT = 0, the absolute (and therefore apparent) magnitudes are too bright by a factor $-2.5 \log_{10} \left(1 + (1-B/T)*10^{0.12041\,\rm{extinct}}\right)$ with respect to the LF defined by the Schechter parameters.

In other words, I would have thought that the 2 following configs should lead to the same LF (the one defined by LF_PHISTAR, LF_MSTAR, and LF_ALPHA), or equivalently to the same apparent magnitude distribution, but it is not the case:


Files

counts.png (24.9 KB) counts.png Damien Le Borgne, 10/07/2016 02:19 PM
Mabs.png (85.3 KB) Mabs.png Damien Le Borgne, 10/18/2016 01:39 PM
Actions #1

Updated by Damien Le Borgne about 5 years ago

stuff replace $10^{0.12041\,\rm{extinct}}$ by $1.0-10^{-0.12041\, \rm{extinct}}$ seems to work.

Actions #2

Updated by Emmanuel Bertin about 5 years ago

  • Status changed from New to In Progress
  • % Done changed from 0 to 30

The whole correction for disk extinction computation definitely needs to be revisited, based on new studies (e.g., SDSS). However for now, let's just keep the De Vaucouleurs et al. (1991) relation between the extinction in the blue $A_B$ and the disk inclination with respect to the line-of-sight $i$:
$$A_B = - \alpha(T)\log \cos i,$$
where $\alpha(T)$ is a type-dependent factor which corresponds to the DISK_EXTINCT configuration parameter (see e.g., the last page of Bertin&Arnouts 1996). Averaging over all inclinations we get (see this link)
$$\langle A_B\rangle_i = +\alpha(T) \log 2$$.
Stuff must generate galaxies using a "face-on" luminosity function with a Schechter $M_0^*$, which should be related to the "observed" luminosity function $M^*$ using
$$M_0^* = M^* -2.5\log_{10} \left(B/T + \frac{(1-B/T)}{\rho}\right),$$
where $B/T$ is the average observed bulge-to-total flux ratio for this category of galaxies, and $\rho$ the average ratio between the observed disk flux and the face-on disk flux in the reference passband (which is assumed to be blue):
$$\rho= 10^{-0.4\,A_B} = 10^{-0.12041\,\alpha(T)}.$$
Hence, one should have, I think:
$$M_0^* = M^* -2.5\log_{10} \left(B/T + (1-B/T)10^{0.12041\,\alpha(T)}\right),$$
which does not seem to match the actual formula in the code. Note also that formally what should be averaged over all angles is $M_i^*$, not $A_B(i)$.

Actions #3

Updated by Emmanuel Bertin about 5 years ago

  • % Done changed from 30 to 40

(!) Edited the formulas above to take into account averaging over all inclination angles.

Actions #4

Updated by Damien Le Borgne about 5 years ago

1) in the default stuff config, DISK_EXTINCT 0.0,0.75,1.23,1.47,1.47,1.23

so alpha is positive. Does it mean $A_B$ is negative ????

2) I didn't understand the code this way :
This idea, I think, as you explained to me, is that the mstar should first be shifted to brighter values in order to account for the fact that disks will be obscured later in the code.

Indeed,the bulge flux is unobscured and the disk obscured, so that the total flux variation that account for the extinction is
$\Delta F_{tot}$ = $F_{disk}^{obscured} - F_{disk}^{unobscured} = F_{disk}^{unobscured} ( \exp^{-\tau}-1)$
so that $\Delta F_{tot}/F_{tot} = (D/T)\, ( \exp^{-\tau}-1) = (1-B/T) ( \exp^{-\tau}-1)$, which is $\le 0$.
and $\Delta m = - 2.5 \log_{10} (1 + \Delta F_{tot}/F_{tot}) = -2.5 \log_{10} (1 + (1-B/T) ( \exp^{ - \tau}-1))$, which is $\ge 0$.

So the formula should be


/* Correct M* for disk internal extinction */
galtype->lf->mstar += 2.5*log10(1.0+(1.0-bt)*(DEXP(-0.12041*extinct)-1));

?
Actions #5

Updated by Damien Le Borgne about 5 years ago

From you paper, $A_B = - \alpha(T) \log \cos i$.
You are missing a $\log$ in your formula above.

Actions #6

Updated by Emmanuel Bertin about 5 years ago

  • % Done changed from 40 to 50

Ah yes, thanks! Now at least I get the right 0.12041 factor. In my case I would only need to replace 1.0 with $B/T$.

Actions #7

Updated by Emmanuel Bertin about 5 years ago

  • Parent task set to #727
Actions #8

Updated by Damien Le Borgne about 5 years ago

  • % Done changed from 50 to 60

... and add the "minus" sign in the exponential (your last formula misses it).

Still, I'm not sure to agree with
$$M_0^* = M^* -2.5\log_{10} \left(B/T + (1-B/T)/\rho\right),$$

Why does the bulge ($B/T$, first term) affect M*, since it is not affected by dust?

Also, to be sure, if I measure blindly a LF from the stuff catalog, am I supposed to recover the LF defined in the stuff parameters by the Schechter function?

Actions #9

Updated by Damien Le Borgne about 5 years ago

Sorry, I guess you are right with your formula (including the minus sign of course which I overlooked).

Actions #10

Updated by Damien Le Borgne about 5 years ago

Still, I have a weird case: if $A_B \rightarrow +\infty$ (i.e. complete extinction of the disk), then $\rho \rightarrow 0$ and $M_0^* - M_0 \rightarrow -\infty$.
Is that expected?...

Actions #11

Updated by Emmanuel Bertin about 5 years ago

Still, I have a weird case: if $A_B \rightarrow +\infty$ (i.e. complete extinction of the disk), then $\rho \rightarrow 0$ and $M_0^* - M_0 \rightarrow -\infty$. Is that expected?...

Yes this is because in that case the formula with $\rho$ is no longer applicable, and you get instead
$$M_0^* = M^* -2.5\log_{10} B/T$$

Actions #12

Updated by Damien Le Borgne about 5 years ago

With your formula above patched to stuff, I measure this in 3 stuff catalogs (details below):

with the following common parameters:


IMAGE_WIDTH     16384     
IMAGE_HEIGHT    16384     
PIXEL_SIZE      0.2       
MAG_LIMITS      16.0,28.0 

PASSBAND_REF    couch/Bj        
CALIBSED_REF    AB            
REFDETECT_TYPE  PHOTONS       
PASSBAND_OBS    couch/Bj  
CALIBSED_OBS    AB        
OBSDETECT_TYPE  PHOTONS   

HUBBLE_TYPE     -6.0     
SED_GALAXIES    E     
SEDINDEX_BULGE  1      
SEDINDEX_DISK   1      

LF_PHISTAR      4.95e-2
LF_MSTAR        -19.58
LF_ALPHA        -0.54
LF_MAGLIMITS    -27.0,-13.0 
LF_PHISTAREVOL  0.
LF_MSTAREVOL    0.

and the 3 sets of particular parameters


CATALOG_NAME    bj.list   
BULGE_FRACTION  1.    
DISK_EXTINCT    0.0


CATALOG_NAME    bj2.list   
BULGE_FRACTION  0.2    
DISK_EXTINCT    0.0

CATALOG_NAME    bj3.list   
BULGE_FRACTION  0.2   
DISK_EXTINCT    10.0

Is it normal that bj3 is shifted bright-wards ?

Actions #13

Updated by Emmanuel Bertin about 5 years ago

This must be a side-effect. Did you turn off galaxy luminosity evolution?

Actions #14

Updated by Damien Le Borgne about 5 years ago

Yes, I set the LF evolution parameters to 0 as you can see, to make thinks easier to understand.

Note: this plot was made using $M_0^* = M^* -2.5\log_{10} \left(B/T + (1-B/T)10^{0.12041\,\alpha(T)}\right)$ only, not your simpler formula at large $A_B$.

Also, why you the formula not be valid for large $A_B$? What is the acceptable range of validity then?
Sorry to bother you... I don't understand everything.... but you can explain me next week if you prefer.

Actions #15

Updated by Emmanuel Bertin about 5 years ago

OK I have to check where side effects kick in (most probably boundaries that depend on $M_0^*$ somewhere).

Actions #16

Updated by Emmanuel Bertin about 5 years ago

  • % Done changed from 60 to 70

Actually as this plot shows one cannot approximate $M_0^*-M^*$ satisfactorily for $\alpha(T)>>1$ using $\langle A_B \rangle_i$, especially for high $B/T$'s. So I guess I will have to numerically average $M_0^*(i)-M^*(i)$ over all inclination angles. btw note that $i$ never reaches $\frac{\pi}{2}$ in stuff.

Actions #17

Updated by Damien Le Borgne about 5 years ago

I went through the code and did some tests.

If we stick to reasonnable alpha (extinct) parameters , i.e. extinct<3, I guess the current code (i.e. with $M_0^* = M^* -2.5\log_{10} \left(B/T + (1-B/T)10^{0.12041\,\alpha(T)}\right),$) is more or less good enough.

But, I may have found a new bug with the normalisation of the LF, shown in an extreme way in #726-12.
More precisely, consider this plot below:
  • upper panel = histogram of mabs from the sampling of the face-on Schechter function in Stuff.
  • lower panel = histogram of mabs + an inclination-dependent k-correction. (So it's almost apparent magnitudes except the distance effect is not applied)

The 3 cases are
  • bj : no extinction, bulge only
  • bj2 : 20% bulge, and no exinction for the disk
  • bj3 : 20% bulge, and extinct=3 for the disk

The bug is: The red line (bj3) in the upper plot should not go above the other luminosity functions : only M* changes (I checked in the code that Phistar and alpha are the same).

When extinction is applied to bj3 (lower panel), the magnitudes more or less fall in the expected range, but there are too many galaxies! (Remember the Phi* values are the same for bj, bj2, bj3, so the LFs should only be shifted along the magnitudes axis...)

In Stuff, I guess the problem comes from lf_intschechter, or from the magmin, magmax variables which are used for the normalisation (?). I can't think of anything else...

Emmanuel can you please check?

Actions

Also available in: Atom PDF